NR-Bitstream video quality metrics for SSIM using encoding decisions in AVC and HEVC coded videos

We propose a no-reference compressed video quality model to predict the full-reference SSIM metrics for AVC (Advanced Video Coding, H.264) and HEVC (High Efficiency Video Coding) videos. The model we use is support vector regression (SVR) model. We use only encoding decisions made during motion estimation to perform the prediction, and do not need the information from pixel domain. We show that the Block-Partition-related features have great importance in SSIM prediction, especially for HEVC videos, due to its partition decisions being more complex than those of AVC. The proposed SVR model trained by data of two different encoding configurations can predict SSIM well for AVC videos (0.78 correlation) and for HEVC videos (0.88 correlation). The proposed models are also compared with a state-of-the-art no-reference-bitstream-pixel SSIM prediction model. We show that the proposed methods provide higher prediction correlation (as high as 13.13% improvement in correlation) with much lower complexity.